Warning method and vehicle

Abstract

This application relates to a warning method and vehicle, specifically in the field of intelligent cockpit technology. The method includes: upon detecting a takeover warning signal, determining the remaining takeover duration, target motion entropy, and takeover operation inertia index, wherein the target motion entropy indicates the degree of change in the trajectory of a road risk object, and the takeover operation inertia index indicates the difficulty of maneuvering the target vehicle to perform an avoidance maneuver; generating a target risk level for the target vehicle based on the remaining takeover duration, target motion entropy, and takeover operation inertia index, wherein the target risk level indicates the degree of driving danger faced by the target vehicle before the driver completes the takeover; and generating corresponding warning information based on the target risk level, wherein the warning information enhances the driver's perception of driving risks. This application improves the safety of driver takeover of vehicles in complex road conditions.

Description

Technical Field

[0001] This application relates to the field of smart cockpit technology, and more particularly to a warning method and vehicle. Background Technology

[0002] With the large-scale production and deployment of autonomous driving technology, autonomous driving takeover requests, as a safety guarantee mechanism in the human-machine co-driving phase, are directly related to whether the driver can take over the vehicle in a timely and safe manner.

[0003] Currently, mainstream early warning schemes rely solely on the duration of a single takeover operation as the basis for triggering a takeover alert. When faced with diverse and complex driving conditions, the risk assessment is one-sided, and the intensity of the alert does not match the actual level of danger in the scenario. This can easily lead to driving safety hazards and is difficult to adapt to the high reliability of takeover safety early warnings required under various complex road conditions. Summary of the Invention

[0004] This application provides a warning method and vehicle to achieve highly reliable takeover safety warnings under complex road conditions.

[0005] Firstly, this application provides a warning method, the method comprising: Upon detecting a takeover warning signal, the remaining takeover duration, target motion entropy, and takeover operation inertia index are determined. The target motion entropy is used to indicate the degree of change in the trajectory of the road risk object, and the takeover operation inertia index is used to indicate the difficulty of maneuvering the target vehicle to perform an avoidance maneuver. Based on the remaining takeover time, the target motion entropy, and the takeover operation inertia index, a target risk level for the target vehicle is generated, wherein the target risk level is used to indicate the degree of driving danger faced by the target vehicle before the driver completes the takeover. Based on the target risk level, a corresponding warning message is generated, wherein the warning message is used to enhance the driver's perception of driving risks.

[0006] Optionally, determining the remaining takeover duration includes: Acquire the initial takeover duration and driver physiological characteristics, wherein the physiological characteristics are used to indicate the driver's concentration level while driving; The time reduction factor is determined based on the driver's physiological characteristics; The remaining takeover duration is obtained by subtracting the time reduction factor from the initial takeover duration.

[0007] Optionally, the driver's physiological characteristics include visual deviation characteristics, fatigue characteristics, and light adaptation characteristics, and determining the time reduction factor based on the driver's physiological characteristics includes: A line-of-sight reduction factor is generated based on the line-of-sight deviation feature, wherein the line-of-sight deviation feature is used to indicate the degree of spatial offset of the driver's line of sight from the vehicle screen; A cognitive arousal reduction factor is generated based on the fatigue characteristics described. A light adaptation reduction factor is generated based on the light adaptation feature, wherein the light adaptation feature is used to indicate the pupil brightness adjustment delay caused by the difference between ambient light and vehicle screen brightness; The time reduction factor is determined based on the sum of the line-of-sight reduction factor, the cognitive arousal reduction factor, and the light adaptation reduction factor.

[0008] Optionally, determining the target motion entropy includes: Obtain the frequency of abrupt changes in the lateral sway angular velocity and longitudinal acceleration of the road risk object; Determine the variance of the lateral oscillation angular velocity; The sum of the variance of the lateral swing angular velocity and the frequency of abrupt changes in the longitudinal acceleration is normalized to generate the target motion entropy.

[0009] Optionally, determining the inertia index of the takeover operation includes: Obtain the equivalent moment of inertia of the steering wheel and the road surface adhesion coefficient; The quotient of the equivalent rotational inertia of the steering wheel and the road surface adhesion coefficient is normalized to generate the inertia index of the takeover operation.

[0010] Optionally, the target risk level of the target vehicle is generated based on the remaining takeover time, the target motion entropy, and the takeover operation inertia index, including: Input the remaining takeover time, the target motion entropy, and the takeover operation inertia index into the risk prediction model; The remaining takeover duration, the target motion entropy, and the takeover operation inertia index are mapped to the same feature space through a feature encoding layer. The risk probability distribution is obtained by weighted summation of various mapped features through a hidden layer. From the risk probability distribution, the risk level corresponding to the highest risk probability is selected as the target risk level.

[0011] Optionally, the risk probability distribution is obtained by weighted summation of the mapped features through a hidden layer, including: Determine the basic weights corresponding to the three types of features: the remaining takeover time, the target motion entropy, and the takeover operation inertia index. Multiple frames of sensor data corresponding to each feature are collected respectively, and the confidence correction coefficient corresponding to each feature is determined based on the fluctuation of the multiple frames of sensor data. Based on the basic weights and confidence coefficients corresponding to each feature, the target weights of each feature are obtained; The risk probability distribution is obtained by weighting and summing the mapped features using the target weights corresponding to each feature.

[0012] Optionally, generating corresponding warning information based on the target risk level includes: Determine the display control parameters corresponding to the target risk level; The display control parameters are sent to the display driver control terminal so that the display driver control terminal can reconstruct the image quality of the vehicle screen based on the written display control parameters. The transmission link for sending the display control parameters bypasses the graphics rendering pipeline of the vehicle operating system.

[0013] Optionally, determining the display control parameters corresponding to the target risk level includes: If the target risk level is the lowest risk level, then the grayscale parameter is determined, wherein the grayscale parameter is used to improve the contrast of the image edges; If the target risk level is an intermediate risk level, then the driving current parameter is determined, wherein the driving current parameter is used to increase the backlight brightness of the area where the road risk object is located and reduce the background backlight, so as to form a contrast between light and dark to enhance the reminder. If the target risk level is the highest risk level, then the color transformation matrix parameters and the masking memory parameters are determined. The color transformation matrix parameters are used to output a single-color warning image, and the masking memory parameters are used to generate a visual funnel containing the road risk object.

[0014] Secondly, this application provides a warning device, the device comprising: The determination module is used to determine the remaining takeover time, target motion entropy, and takeover operation inertia index when a takeover warning signal is detected. The target motion entropy is used to indicate the degree of change in the trajectory of the road risk object, and the takeover operation inertia index is used to indicate the difficulty of the target vehicle to perform an avoidance operation. The first generation module is used to generate a target risk level of the target vehicle based on the remaining takeover time, the target motion entropy, and the takeover operation inertia index, wherein the target risk level is used to indicate the degree of driving danger faced by the target vehicle before the driver completes the takeover. The second generation module is used to generate corresponding warning information based on the target risk level, wherein the warning information is used to enhance the driver's perception of driving risks.

[0015] Thirdly, this application provides an electronic device, comprising: at least one communication interface; at least one bus connected to the at least one communication interface; at least one processor connected to the at least one bus; and at least one memory connected to the at least one bus.

[0016] Fourthly, this application also provides a vehicle, the vehicle including a memory and a processor, wherein the memory stores executable program code, and the processor is used to call and execute the executable program code to implement the method described in any of the above.

[0017] The technical solutions provided in this application have the following advantages compared with the prior art: In this application, upon detecting a takeover warning signal, the remaining takeover duration, target motion entropy, and takeover operation inertia index are obtained. The remaining takeover duration reflects the available time for the driver to perform hazard avoidance maneuvers, the target motion entropy reflects the suddenness of obstacle movement, and the takeover operation inertia index reflects the difficulty of vehicle avoidance maneuvers. By combining these three types of indicators, the actual driving risks can be reconstructed, the differences in the degree of danger under various road conditions can be accurately distinguished, and the accuracy of the target risk level judgment results can be improved. The warning information output based on the target risk classification can also match the actual degree of danger in the scenario, allowing the driver to intuitively perceive the current driving hazards and complete the vehicle takeover in a timely manner, reducing the occurrence of collision accidents and improving the safety of the driver taking over the vehicle in complex road conditions. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0021] Figure 1 A schematic diagram of the warning system provided in the embodiments of this application; Figure 2 A flowchart of a warning method provided in an embodiment of this application; Figure 3This is a schematic diagram of the structure of a warning device provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] The following disclosure provides numerous different embodiments or examples for implementing various structures of this application. To simplify the disclosure, specific examples of components and arrangements are described below. These are merely examples and are not intended to limit the scope of this application. Furthermore, reference numerals and / or letters may be repeated in different examples. Such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0024] The warning method provided in this application embodiment can be applied to processing scenarios where the vehicle processor runs independently, as well as to hardware environments where the vehicle and server process collaboratively. In collaborative processing scenarios, such as Figure 1 As shown, server 103 is connected to vehicle 101 via a network. Vehicle 101 sends a takeover warning signal to server 103. Server 103 generates a target risk level based on the takeover warning signal and feeds it back to vehicle 101. Vehicle 101 generates corresponding warning information according to the target risk level. Database 105 can be set up on the server or independently of the server to provide data storage services for server 103. The above network includes, but is not limited to, wide area network, metropolitan area network or local area network.

[0025] Furthermore, the embodiments of this application can also be implemented as a functional plug-in, integrated into the vehicle processor, and directly generate warning information based on the takeover warning signal.

[0026] The following will describe in detail a warning method provided in this application embodiment, taking a processor applied to a vehicle as an example, with specific implementation details. Figure 2 As shown, the specific steps are as follows: Step 201: Upon detecting a takeover warning signal, determine the remaining takeover duration, target motion entropy, and takeover operation inertia index. The target motion entropy is used to indicate the degree of change in the trajectory of the road risk object, and the takeover operation inertia index is used to indicate the difficulty of maneuvering the target vehicle to perform an avoidance maneuver. Step 202: Based on the remaining takeover time, target motion entropy, and takeover operation inertia index, generate the target risk level of the target vehicle. The target risk level is used to indicate the degree of driving danger faced by the target vehicle before the driver completes the takeover. Step 203: Generate corresponding warning information based on the target risk level. The warning information is used to enhance the driver's perception of driving risks.

[0027] In step 201, the processor acquires the driving perception data collected by the vehicle in real time. When it is determined that there is a potential collision risk ahead and the vehicle cannot continue to drive autonomously and safely, it will actively trigger a takeover warning signal. After the processor recognizes the valid takeover warning signal, it will simultaneously acquire three types of indicators: the remaining takeover time, the target motion entropy, and the takeover operation inertia index.

[0028] The remaining takeover time represents the total available time for the driver to take evasive action if the vehicle continues to travel at its current speed until a collision occurs. The smaller the remaining takeover time value, the less time the driver has to take evasive action.

[0029] Target motion entropy refers to the degree of drastic change in the trajectory of a road risk object. The higher the target motion entropy value, the more variable the movement behavior of the road risk object, and the higher the risk of sudden collision. For example, a vehicle traveling straight at a constant speed has a lower motion entropy value, while a pedestrian suddenly turning back or an electric vehicle weaving has a higher motion entropy value. Road risk objects refer to road targets detected by radar and cameras that have the potential to collide with a target vehicle; these can be moving targets such as vehicles, pedestrians, and electric vehicles.

[0030] The takeover inertia index refers to the difficulty of maneuvering a target vehicle to perform an avoidance maneuver. The higher the takeover inertia index, the greater the difficulty for the driver to complete the avoidance maneuver. The takeover inertia index is affected by factors such as vehicle speed, vehicle weight, and road surface wetness. For example, the takeover inertia index is lower when driving on dry paved roads, while it will increase significantly when driving at high speeds on wet or slippery roads.

[0031] In step 202, the processor retrieves the remaining takeover time, target motion entropy, and takeover operation inertia index, and combines these three indicators to generate the target risk level of the target vehicle. The target risk level is a standardized level that the processor classifies as the degree of danger of the current driving scenario. The higher the target risk level, the more urgent the collision risk that the vehicle faces before the driver takes over.

[0032] The remaining takeover time reflects the available time for hazard avoidance, the target motion entropy reflects the suddenness of obstacle movement, and the takeover operation inertia index reflects the difficulty of vehicle control. The processor fully restores the real driving risks from three dimensions: available time for hazard avoidance, suddenness of obstacle movement, and difficulty of vehicle control. It accurately distinguishes the differences in real dangers under different working conditions and avoids inaccurate risk level assessment caused by a single indicator. In addition, the three types of indicators are mutually verified, and short-term data fluctuations of a single indicator will not have a significant impact on the risk level determination result. The stability of the final output target risk level is also stronger.

[0033] For example, the remaining takeover time for each scenario is 4 seconds. In the first scenario, the vehicle in front is traveling straight at a constant speed and the road surface is dry, which is judged as a low-risk level. In the second scenario, a pedestrian suddenly crosses the road and the road surface is slippery. The target motion entropy and the takeover operation inertia index will increase simultaneously, which is judged as a high-risk level.

[0034] In step 203, the processor generates warning information adapted to the current level of danger based on the target risk level. This warning information includes various interactive forms, such as tiered voice alarms output from the vehicle's speakers or visual reminders created by adjusting the display on the central control screen. It can be output individually or in combination, its purpose being to amplify risk perception, remind the driver to take over the vehicle in a timely manner, avoid collisions, and improve driving safety.

[0035] In this application, upon detecting a takeover warning signal, the remaining takeover duration, target motion entropy, and takeover operation inertia index are obtained. The remaining takeover duration reflects the available time for the driver to perform hazard avoidance maneuvers, the target motion entropy reflects the suddenness of obstacle movement, and the takeover operation inertia index reflects the difficulty of vehicle avoidance maneuvers. By combining these three types of indicators, the actual driving risks can be reconstructed, the differences in the degree of danger under various road conditions can be accurately distinguished, and the accuracy of the target risk level judgment results can be improved. The warning information output based on the target risk classification can also match the actual degree of danger in the scenario, allowing the driver to intuitively perceive the current driving hazards and complete the vehicle takeover in a timely manner, reducing the occurrence of collision accidents and improving the safety of the driver taking over the vehicle in complex road conditions.

[0036] As an optional implementation, step 201, determining the remaining takeover time, includes: Acquire the initial takeover duration and driver physiological characteristics, whereby the physiological characteristics are used to indicate the driver's level of concentration while driving; Determine the time reduction factor based on the driver's physiological characteristics; The remaining takeover time is obtained by subtracting the time reduction factor from the initial takeover time.

[0037] The processor uses vehicle radar and cameras to collect the relative distance and relative speed between its target vehicle and road hazard objects, and then calculates the initial takeover time. For example, if the relative distance between the two vehicles is 30 meters and they are moving closer to each other at a rate of 10 meters per second, the remaining takeover time is calculated to be 3 seconds.

[0038] The initial takeover time is derived solely from vehicle motion and represents the theoretical time available for the driver to avoid hazards under ideal conditions. However, this theoretical value has significant limitations, as the calculation process completely disregards the impact of the driver's real-time state on reaction speed. In actual driving, it is difficult for drivers to maintain high levels of concentration at all times: a driver's gaze shifting to the side window, prolonged periods of looking down at in-vehicle equipment, drowsiness, or significant differences in light levels inside and outside the vehicle causing temporary blurred vision and squinting can all prolong the time it takes for the driver to detect road hazards and take over. In such situations, even if the initial takeover time is sufficient, the driver may not be able to complete the hazard avoidance maneuver within the theoretical timeframe, easily leading to insufficient reaction time and delayed avoidance, posing a significant safety hazard.

[0039] Therefore, the processor synchronously retrieves driver physiological characteristic data collected by in-vehicle sensing devices. These physiological characteristics are specifically categorized into gaze deviation characteristics, fatigue characteristics, and light adaptation characteristics. Gait deviation characteristics indicate whether the driver's eyes are continuously focused on the road ahead; fatigue characteristics reflect the driver's level of alertness; and light adaptation characteristics reflect the visual delay when the human eye switches between light and dark environments. These three types of physiological characteristics, from the dimensions of visual attention, mental state, and visual adaptation, comprehensively reflect the driver's current level of focus, thus providing an overall indication of the driver's level of concentration while driving.

[0040] The processor integrates the detection results of three types of physiological characteristics and performs unified calculations to generate corresponding time reduction factors. The time reduction factor is a correction value used to offset the initial takeover time, that is, to correct the theoretical risk avoidance time. The more obvious the driver's adverse state, such as distraction, drowsiness, or visual adaptation delay, the larger the calculated time reduction factor value, which means that the driver needs to spend more time to complete the takeover operation.

[0041] The processor subtracts a time reduction factor from the initial takeover time to output the remaining takeover time. By introducing a time reduction factor to correct the theoretical time, it can compensate for the time deviation caused by relying solely on vehicle motion data for calculation, ensuring that the final remaining takeover time matches the driver's actual reaction ability at that moment.

[0042] For example, the vehicle calculates an initial takeover time of 5 seconds using radar and cameras. However, if the driver looks down at their phone and their gaze is off the road for an extended period, the time reduction factor is calculated to be 2 seconds, resulting in a remaining takeover time of 3 seconds. This value is closer to the actual amount of time the driver can use to avoid danger in their current state, allowing subsequent risk level assessments to better reflect real-world driving scenarios.

[0043] The formula for calculating the remaining takeover time is: TTR_Dynamic=TTR_ADAS T_comp Where TTR_Dynamic is the remaining takeover duration, TTR_ADAS is the initial takeover duration, and T_comp is the time reduction factor.

[0044] In this application, a time reduction factor is generated by collecting multi-dimensional physiological characteristics of the driver to correct the initial takeover time calculated solely based on vehicle motion data. Instead of using an idealized driver state as the risk assessment benchmark, this approach fully considers various real-world driving conditions such as driver distraction, fatigue, and visual light-dark adaptation delays. This reduces the difference between the theoretical avoidance time and the driver's actual available reaction time, preventing situations where the system misjudges sufficient risk when the driver's actual reaction is insufficient. This makes subsequent risk level determination based on the remaining takeover time more accurate, and the corresponding warning information can also match the driver's actual reaction shortcomings in advance, reminding the driver to take over the vehicle earlier. This reduces the collision risk caused by poor driver condition from the source and further improves driving safety in complex human-machine co-driving scenarios.

[0045] As an optional implementation, the driver's physiological characteristics include visual deviation characteristics, fatigue characteristics, and light adaptation characteristics. Determining the time reduction factor based on these driver physiological characteristics includes: A line-of-sight reduction factor is generated based on line-of-sight deviation features, which are used to indicate the degree of spatial offset of the driver's line of sight from the vehicle's infotainment screen. Generate a cognitive arousal reduction factor based on fatigue characteristics; A light adaptation reduction factor is generated based on light adaptation characteristics, whereby the light adaptation characteristics are used to indicate the pupil dimming delay caused by the difference between ambient light and the brightness of the vehicle screen. The time reduction factor is determined based on the sum of the visual reduction factor, cognitive arousal reduction factor, and light adaptation reduction factor.

[0046] Driver physiological characteristics include line-of-sight deviation, fatigue, and light adaptation. The processor calculates individual reduction factors for each type of characteristic and adds these three types of individual reduction factors to obtain a unified time reduction factor. The generation process of each reduction factor is shown below.

[0047] Based on the gaze deviation feature, the processor calculates a gaze reduction factor. The gaze deviation feature represents the spatial offset of the driver's gaze point relative to the risk warning area on the vehicle's infotainment screen. The processor identifies the driver's gaze coordinates and the two-dimensional coordinates of the risk area on the screen, then calculates the spatial offset angle between the two coordinates. The size of this angle directly corresponds to the degree of spatial offset represented by the gaze deviation feature. A larger spatial offset angle means the driver needs to rotate their eyes and head to see the risk warning on the screen, resulting in a longer gaze shift and a correspondingly higher gaze reduction factor value.

[0048] For example, when the driver's gaze is completely directed out of the window, the spatial offset angle is larger and the line-of-sight reduction factor is higher; when the driver's gaze is directed at the central control screen, the spatial offset angle is extremely small and the line-of-sight reduction factor is close to 0.

[0049] Based on fatigue characteristics, the processor calculates a cognitive arousal reduction factor. Fatigue characteristics are categorized into different fatigue levels by statistically analyzing the duration and percentage of time the driver keeps their eyes closed, thus determining the driver's level of alertness. When the driver is fully awake and focused, the cognitive arousal reduction factor is 0. When the driver experiences frequent eye closures, drowsiness, or other signs of fatigue, the brain's ability to receive risk signals and make operational decisions slows down, and the cognitive arousal reduction factor dynamically increases. This factor can be set to a maximum of 1.2 seconds.

[0050] Regarding light adaptation features, the processor calculates a light adaptation reduction factor based on these features. Light adaptation features reflect the visual recognition delay caused by the contraction and dilation of the human pupil when there is a difference between the ambient light intensity outside the vehicle and the brightness of the vehicle's infotainment screen. The processor compares the ambient light intensity outside the vehicle with the brightness of the vehicle's infotainment screen to obtain the light difference. The larger the light difference, the longer the pupil takes to adjust its brightness, and the light adaptation reduction factor increases accordingly. For example, the reduction factor ranges from 0.1 seconds to 0.4 seconds. For instance, at night, when the ambient light outside the vehicle is dim and the screen brightness is high, the brightness difference is large, and the light adaptation reduction factor is 0.4 seconds; during the day, when the outdoor light is strong and the screen brightness is increased, the brightness difference is small, and the light adaptation reduction factor is only 0.1 seconds.

[0051] After the processor calculates the gaze reduction factor, cognitive arousal reduction factor, and light adaptation reduction factor respectively, it adds the three types of individual reduction factors together, and the sum is the time reduction factor.

[0052] This application quantifies three physiological characteristics affecting driver reaction speed—visual deviation, mental fatigue, and visual delay due to varying light conditions—separately, generating corresponding individual time reduction factors for each. These factors are then combined to obtain the overall time reduction factor. This allows for a more precise breakdown of reaction time loss corresponding to each driver's adverse state. Compared to a general estimation of the overall reduction time, this method provides a more accurate measurement of driver reaction delay. Furthermore, the three reduction factors are calculated independently. Any deterioration in driver condition in any dimension will individually increase the corresponding reduction factor, ensuring that the avoidance time loss caused by fatigue, visual deviation, or lighting differences is not overlooked. This allows the time reduction factor to accurately reflect the driver's overall reaction delay, thereby correcting the remaining takeover time to match the driver's real-time physiological state and avoiding the bias of generating risk levels solely based on vehicle motion data.

[0053] As an optional implementation, in step 201, determining the target motion entropy includes: obtaining the lateral sway angular velocity and longitudinal acceleration mutation frequency of the road risk object; determining the lateral sway angular velocity variance; and normalizing the sum of the lateral sway angular velocity variance and the longitudinal acceleration mutation frequency to generate the target motion entropy.

[0054] The processor obtains the lateral sway angular velocity and longitudinal acceleration change frequency of road risk objects through sensing devices.

[0055] Among them, the lateral oscillation angular velocity is used to reflect the severity of left and right lane changes, turns, or lateral deviations of road risk objects. In order to eliminate the interference caused by instantaneous jitter data, the processor will calculate the variance corresponding to the angular velocity. The higher the variance value, the more unstable the lateral movement and the stronger the lateral sudden risk.

[0056] The frequency of longitudinal acceleration abrupt changes is used to statistically determine the frequency of changes in the forward and backward motion of road risk objects during acceleration, deceleration, and sudden braking. The higher the frequency of abrupt changes, the greater the fluctuation in the longitudinal driving state, and the more likely it is to cause dangerous situations such as sudden deceleration.

[0057] The processor simultaneously uses the variance of lateral oscillation angular velocity and the frequency of longitudinal acceleration abrupt changes to calculate the target motion entropy. The two methods work together to comprehensively cover both lateral offset and longitudinal acceleration / deceleration changes of road risk objects, capturing all kinds of sudden dangerous actions from these objects. Using only single-dimensional data creates blind spots in the assessment. Lateral data alone cannot identify longitudinal risks such as frequent sudden braking by vehicles ahead, and longitudinal data alone cannot determine lateral risks such as pedestrians crossing the road or vehicles changing lanes continuously. Combining the two dimensions of data avoids the omission of motion risks caused by a single dimension, allowing the target motion entropy to comprehensively and objectively reflect the severity of changes in the trajectory of road risk objects.

[0058] The processor adds the variance of the lateral oscillation angular velocity to the frequency of abrupt changes in longitudinal acceleration to obtain a comprehensive value representing the degree of motion turbulence of the road risk object. This comprehensive value is then normalized to map the values ​​uniformly to the [0,1] interval, ultimately generating a standardized target motion entropy. Normalization eliminates the inconsistency in units and magnitudes between the two types of parameters, allowing the target motion entropy to be used in conjunction with the remaining takeover duration and the takeover operation inertia index in subsequent risk level calculations.

[0059] For example, a motor vehicle traveling at a constant speed in a straight line has a lateral angular velocity variance close to 0 and a very low frequency of longitudinal acceleration abrupt changes, resulting in a low target motion entropy value after summing the two. A pedestrian who repeatedly crosses the lane has a large lateral angular velocity variance and frequent starts and stops bring a large number of longitudinal acceleration abrupt changes, resulting in a high target motion entropy value after summing the two, accurately reflecting the pedestrian's high risk of sudden events.

[0060] The formula for calculating the target motion entropy is as follows: E_entropy=norm(σ_lat²+f_along), where E_entropy is the target motion entropy, σ_lat² is the variance of the lateral swing angular velocity, and f_along is the frequency of longitudinal acceleration abrupt changes.

[0061] In this application, motion characteristics of road risk objects are extracted from both lateral and longitudinal motion dimensions. The variance of the lateral angular velocity represents the severity of lateral displacement, while the frequency of acceleration abrupt changes quantifies longitudinal acceleration and deceleration fluctuations. These two parameters work together to comprehensively capture various sudden dangerous movements of road risk objects, avoiding missed detections of motion risks caused by a single dimension. The target motion entropy generated by summing the two parameters integrates bidirectional motion turbulence information, enabling an objective and comprehensive quantification of the suddenness of obstacle motion, which is beneficial to the accuracy of subsequent target risk level determination.

[0062] As an optional implementation, in step 201, determining the takeover operation inertia index includes: obtaining the equivalent rotational inertia of the steering wheel and the road surface adhesion coefficient; normalizing the quotient of the equivalent rotational inertia of the steering wheel and the road surface adhesion coefficient to generate the takeover operation inertia index.

[0063] The processor reads the chassis operating data and extracts the equivalent rotational inertia of the steering wheel and the road surface adhesion coefficient.

[0064] The equivalent moment of inertia of the steering wheel reflects the steering resistance caused by the vehicle's weight and speed. The heavier the vehicle and the higher the speed, the greater the equivalent moment of inertia of the steering wheel, and the more effort the driver needs to exert to turn the steering wheel to adjust the driving trajectory.

[0065] The coefficient of friction reflects a tire's ability to grip the road surface. The drier the road surface, the higher the coefficient of friction. The coefficient of friction will decrease significantly on rainy, snowy, or icy roads. Vehicles are more likely to slip on low-friction surfaces, making avoidance maneuvers much more difficult.

[0066] The processor divides the equivalent moment of inertia of the steering wheel by the road surface adhesion coefficient to obtain a quotient, then performs a normalization operation on this quotient, uniformly mapping it to the [0,1] interval, ultimately generating a standardized takeover inertia index. Since the equivalent moment of inertia of the steering wheel only reflects the vehicle's own steering load and cannot reflect the driving limitations caused by road surface grip conditions, and relying solely on the road surface adhesion coefficient can only determine the risk of road slippage, it cannot reflect the vehicle's own steering resistance. Calculating these two types of parameters together can simultaneously consider both the vehicle's own handling load and road driving conditions, fully restoring the physical handling costs that the driver faces when avoiding other vehicles, and avoiding the biased judgment problems caused by a single parameter evaluation.

[0067] The higher the inertia index of the takeover operation, the more difficult the vehicle's avoidance maneuver under the current operating conditions. High-speed driving, slippery road surfaces, and heavy vehicle loads all increase this index value. For example, on dry asphalt roads and when driving at low speeds with no load, the equivalent rotational inertia of the steering wheel is small and the road surface adhesion coefficient is high, resulting in a small inertia index of the takeover operation. On wet or snowy roads and when driving at high speeds with heavy loads, the equivalent rotational inertia of the steering wheel increases significantly, the road surface adhesion coefficient decreases, and the inertia index of the takeover operation is high.

[0068] The formula for calculating the inertia index of the takeover operation is as follows: I_inertia=norm(J_eps / μ), where I_inertia is the inertia index of the takeover operation, J_eps is the equivalent rotational inertia of the steering wheel, and μ is the road surface adhesion coefficient.

[0069] This application combines the equivalent rotational inertia of the steering wheel and the road surface adhesion coefficient to jointly calculate the takeover operation inertia index. This simultaneously covers two key factors affecting the difficulty of obstacle avoidance: the vehicle's own steering resistance and the road surface grip capability. This overcomes the limitations of evaluating driving conditions with a single parameter and can objectively quantify the real handling costs under different vehicle speeds, loads, and road conditions. Furthermore, by normalizing the quotient to transform it into a takeover operation inertia index with a unified scale, it can be used in conjunction with the remaining takeover time and the target motion entropy to participate in the risk level calculation. This fully restores the three risk dimensions of available avoidance time, sudden risks, and vehicle handling difficulty, effectively distinguishing the real danger levels under different road conditions. It avoids the problem of underestimating risks caused by ignoring vehicle handling difficulty, making subsequent risk assessment and graded warnings more consistent with real driving conditions and improving the overall driving safety redundancy during the autonomous driving takeover phase.

[0070] As an optional implementation, in step 202, generating the target risk level of the target vehicle based on the remaining takeover time, target motion entropy, and takeover operation inertia index includes: inputting the remaining takeover time, target motion entropy, and takeover operation inertia index into the risk prediction model; mapping the remaining takeover time, target motion entropy, and takeover operation inertia index to the same feature space through a feature encoding layer; performing a weighted summation of the mapped features through a hidden layer to obtain a risk probability distribution; and selecting the risk level corresponding to the highest risk probability from the risk probability distribution as the target risk level.

[0071] The risk prediction model has a three-layer structure: an input layer, a hidden layer, and an output layer. The input layer receives three-dimensional feature vectors: the remaining takeover time (TTR_Dynamic), the target motion entropy (E_entropy), and the takeover operation inertia index (I_inertia). These three feature vectors correspond to the available time for risk avoidance, the degree of suddenness of obstacle movement, and the difficulty of vehicle avoidance maneuvers, respectively.

[0072] Since the numerical ranges and dimensions of the three types of feature vectors are not consistent, they cannot be directly used in unified calculations. Therefore, risk prediction uses a feature encoding layer to perform a unified transformation on the input three-dimensional features, mapping the three types of feature vectors to the same feature space and eliminating the computational interference caused by the differences in numerical ranges between different indicators.

[0073] After feature mapping is completed, deep feature fusion is performed through a hidden layer. The hidden layer has 6 neurons and uses activation functions, such as the ReLU function, to uncover complex nonlinear coupling relationships between features. The hidden layer assigns specific weights to the three types of encoded feature vectors, performs a weighted summation operation, and explores the correlation between the three types of feature vectors, eliminating judgment bias caused by short-term fluctuations of a single feature.

[0074] The data processed in the hidden layer is fed into the output layer. The output layer uses a multi-probability activation function, such as the Softmax function, to output a risk probability distribution corresponding to three levels of risk urgency. The three probability values ​​are P1, P2, and P3, which are respectively associated with low risk, medium risk, and high risk levels. The processor compares the values ​​of the three risk probabilities and selects the risk level corresponding to the highest risk probability value, which is then determined as the target risk level for the current scene.

[0075] For example, the vehicle calculates that the remaining takeover time is 4.5 seconds. The vehicle ahead is moving straight at a constant speed with small lateral sway and smooth acceleration and deceleration, and the target motion entropy is 0.4. The current vehicle is traveling on dry asphalt but is fully loaded, and the takeover operation inertia index is 0.6. The three sets of data are input into the risk prediction model. After feature encoding and weighted operation in the hidden layer, the output layer gives the probability distribution corresponding to the three risk levels: P1 = 0.82 for low risk, P2 = 0.15 for medium risk, and P3 = 0.03 for high risk. Among the three probabilities, P1 has the largest value, and P1 corresponds to low risk. Therefore, the processor determines that the scenario is of low risk level. For example, if the vehicle has 2.2 seconds of remaining takeover time, a pedestrian suddenly crosses the lane, frequently changes speed, and the target's motion entropy is 0.8, while the road surface is slippery due to rain or snow and the vehicle is traveling at high speed and heavily loaded, resulting in a takeover operation inertia index of 0.9, all three data points indicate a high level of danger in the scenario. After the data is input into the risk prediction model, the output probability distribution is P1=0.02, P2=0.18, and P3=0.80. Among the three probability groups, P3 has the highest value, corresponding to high risk. The processor determines that the scenario is of a high-risk level.

[0076] In this application, feature vectors of various dimensions are first unified through feature encoding to solve the problem of inconsistent scales of different risk features, which prevent joint computation. Then, the feature vectors are weighted and fused by a multi-neuron hidden layer and activation function to fully explore the correlation between avoidance time, the degree of suddenness of obstacle movement, and the difficulty of vehicle control, thereby weakening the misjudgment caused by instantaneous anomalies in a single data point. Finally, standardized multi-level risk probabilities are output, and the risk level is divided by taking the highest probability to achieve a non-linear comprehensive evaluation of multi-dimensional risk levels. The target risk level is closely aligned with the actual danger level of the scenario.

[0077] As an optional implementation, the risk probability distribution is obtained by weighted summation of the mapped features through a hidden layer, including: Determine the basic weights corresponding to the three types of features: the remaining takeover time, the target motion entropy, and the takeover operation inertia index. Multiple frames of sensor data corresponding to each feature are collected respectively, and the confidence correction coefficient corresponding to each feature is determined based on the fluctuation of the multiple frames of sensor data. Based on the basic weights and confidence coefficients corresponding to each feature, the target weights of each feature are obtained; The risk probability distribution is obtained by weighting and summing the mapped features using the target weights corresponding to each feature.

[0078] The processor retrieves the pre-stored weight parameters within the risk prediction model and determines the basic weights corresponding to the three types of features: remaining takeover time, target motion entropy, and takeover operation inertia index. These basic weights characterize the contribution of each type of feature to the overall risk assessment.

[0079] The processor collects multiple frames of sensor data corresponding to the three types of features respectively, and determines the confidence correction coefficient corresponding to each feature based on the fluctuation of the multiple frames of sensor data.

[0080] Regarding the remaining takeover time, the processor continuously collects the relative distance and relative speed between the target vehicle and road risk objects across multiple frames, using this as multi-frame sensor data. Then, it statistically analyzes the fluctuation amplitude of the relative distance and relative speed across these frames, using this as a fluctuation assessment index. Since larger fluctuations in relative distance and relative speed indicate distortion of the radar sensing signal and lower reliability of the calculated remaining takeover time, the corresponding first confidence correction coefficient decreases accordingly, approaching 0. Conversely, when the relative distance and relative speed are stable without abrupt changes, the first confidence correction coefficient approaches 1. Therefore, the fluctuation degree of the multi-frame sensor data corresponding to the remaining takeover time is negatively correlated with the first confidence correction coefficient.

[0081] To assess the target motion entropy, the processor continuously collects the variance of the lateral angular velocity and the frequency of abrupt changes in the longitudinal acceleration of the road risk object across multiple frames. This data is then used as multi-frame sensing data. The overall fluctuation amplitude of these fluctuations across multiple frames is then statistically analyzed and used as a volatility evaluation index. Since greater fluctuations in the frequency of abrupt changes in lateral angular velocity and longitudinal acceleration indicate unstable visual perception and lower reliability of the calculated target motion entropy, the corresponding second confidence correction coefficient decreases accordingly, approaching 0. Conversely, when the frequency of abrupt changes in lateral angular velocity and longitudinal acceleration is stable without jumps, the second confidence correction coefficient approaches 1. Therefore, the degree of fluctuation in the multi-frame sensing data corresponding to the target motion entropy is negatively correlated with the second confidence correction coefficient.

[0082] For the takeover operation inertia index, the processor continuously collects the equivalent rotational inertia of the steering wheel and the road surface adhesion coefficient across multiple frames, using these as multi-frame sensor data. Then, it statistically analyzes the fluctuation amplitude of the equivalent rotational inertia of the steering wheel and the road surface adhesion coefficient across these frames, using this as a fluctuation evaluation index. Since a larger fluctuation in the equivalent rotational inertia of the steering wheel and the road surface adhesion coefficient indicates distortion of the chassis sensor signal, the calculated takeover operation inertia index has lower reliability. Therefore, the corresponding third confidence correction coefficient decreases accordingly, approaching 0. Conversely, when the equivalent rotational inertia of the steering wheel and the road surface adhesion coefficient are stable without abrupt changes, the third confidence correction coefficient approaches 1. Therefore, the fluctuation degree of the multi-frame sensor data corresponding to the takeover operation inertia index is negatively correlated with the third confidence correction coefficient.

[0083] The processor multiplies the base weights corresponding to each type of feature with the dedicated confidence correction coefficients to obtain the target weights of each feature. Then, the processor uses the corresponding target weights to scale the standardized features in the same feature space output by the feature encoding layer. Finally, the processor sums the three types of features element by element after scaling to complete the multi-risk dimension adaptive fusion operation and outputs the risk probability distribution corresponding to low, medium and high risk levels.

[0084] In this application, dedicated multi-frame sensor data is matched for each of the three types of features. The perception confidence is judged by the fluctuation amplitude of the multi-frame sensor data. The fluctuation degree of the multi-frame sensor data for all features is negatively correlated with the confidence correction coefficient. Then, the target weight of each feature is obtained by multiplying the basic weight with the confidence correction coefficient. This adaptive weighted fusion mechanism abandons the traditional fixed weight direct weighted fusion method, takes into account the contribution of the basic weight and the fluctuation of the multi-frame sensor data, effectively suppresses the risk probability shift caused by short-term sensor disturbances, and improves the stability of the risk probability distribution under complex working conditions.

[0085] As an optional implementation, this application also provides an offline training process for the above-mentioned risk prediction model. The training process is as follows: a large amount of data covering different road surfaces and different road risk targets are collected in batches to construct a training sample set. In addition to sample takeover duration, sample motion entropy, sample takeover operation inertia index, and risk level label, each set of training samples also generates multiple frames of sensor data required for that set of training samples. Among them, the multiple frames of sensor data associated with sample takeover duration are multiple frames of relative distance and relative velocity, the multiple frames of sensor data associated with sample motion entropy are multiple frames of lateral sway angular velocity variance and longitudinal acceleration change frequency, and the multiple frames of sensor data associated with sample takeover operation inertia index are multiple frames of steering wheel equivalent rotational inertia and road surface adhesion coefficient.

[0086] The processor inputs each set of training samples into the initial risk prediction model, which then sequentially receives features through the input layer, maps them uniformly to the same feature space through the feature encoding layer, and then enters a hidden layer with 6 neurons and configured activation functions to perform adaptive weighted fusion. The fusion process first reads the basic weights of the three types of features, then determines the confidence correction coefficients corresponding to each feature based on the fluctuation amplitude of the accompanying multi-frame sensor data. The basic weights are multiplied by the confidence correction coefficients to obtain the target weights of each feature. The target weights are used to scale each feature and then summed. Finally, the output layer outputs the risk probability distribution, and the risk level corresponding to the highest probability is selected as the actual risk level output by the model.

[0087] The processor compares the actual risk level output by the model with the risk level label in real time, calculates the difference in classification loss, and backpropagates based on this difference to iteratively update all network parameters in the feature encoding layer and hidden layer of the model, continuously reducing the judgment deviation between the actual risk level and the risk level label. The entire training process of sample input, feature encoding, adaptive weighted fusion, risk probability output, loss calculation, and network parameter update is executed in a loop, repeatedly optimizing the weight parameters inside the model until the global loss value after multiple iterations stabilizes and converges, at which point training can be stopped, and the trained risk prediction model is obtained.

[0088] In this application, during the offline training phase, multi-frame sensor data corresponding to various features are simultaneously incorporated, allowing the model to fully learn the intrinsic relationship between the stability of sensor signals and the three types of risk features. After training convergence, the solidified basic weights can accurately represent the risk contribution ratio of various features. Then, the fluctuations of multi-frame sensor data are superimposed to generate target weights, which can automatically reduce the impact of distorted sensor data on risk judgment, avoid misjudgment of risk level caused by instantaneous changes in a single sensor, and improve the reliability of risk assessment results in complex takeover scenarios such as mixed traffic of people and vehicles and slippery conditions caused by rain and snow.

[0089] As an optional implementation, step 203, generating corresponding warning information based on the target risk level, includes: determining the display control parameters corresponding to the target risk level; sending the display control parameters to the display driver control terminal so that the display driver control terminal can reconstruct the image quality of the vehicle screen based on the written display control parameters, wherein the transmission link for sending the display control parameters bypasses the graphics rendering pipeline of the vehicle operating system.

[0090] After the processor completes the target risk level determination, it retrieves the pre-stored mapping relationship, which stores the display control parameters corresponding to each risk level. The processor matches the display control parameters specific to the target risk level. The display control parameters are used to adjust at least one of the following screen parameters: overall screen brightness, backlight zone brightness, screen contrast, and high brightness of the warning area. The higher the target risk level, the greater the brightening of the warning screen and the gain of the zone backlight in the display control parameters, in order to enhance the visual reminder effect.

[0091] After the processor generates the display control parameters, it does not transmit the parameters to the vehicle operating system for graphics rendering. Instead, it sends them directly to the display driver control terminal corresponding to the vehicle screen through an independent hardware transmission channel. The entire data transmission link bypasses the graphics rendering pipeline built into the vehicle operating system.

[0092] Because the operating system's graphics rendering pipeline needs to process a large number of conventional screen layers such as navigation, multimedia, and vehicle status, the data processing flow is complex and the forwarding latency is high. If the display control parameters are forwarded through this graphics rendering pipeline, it will lengthen the response time of the screen warning display. In this embodiment, the display driver control terminal directly receives the display control parameters issued by the processor, and adjusts the screen imaging logic in real time according to the values ​​in the display control parameters. This quickly reconstructs the overall image quality of the vehicle screen, specifically enhancing the display effect of the risk warning area, and conveying a visual warning to the driver that matches the target risk level.

[0093] For example, when the processor determines that the current scenario is of low risk level, the matching display control parameters only slightly increase the brightness of the risk warning text; if it is determined to be of high risk level, the display control parameters will simultaneously increase the backlight brightness of the warning area and improve the screen contrast. After receiving the display control parameters, the display driver control terminal immediately reconstructs the screen image quality, highlights the risk warning image, and quickly attracts the driver's attention.

[0094] In this application, when outputting visual warning information, display control parameters are sent through an independent transmission channel, bypassing the graphics rendering pipeline of the vehicle's operating system. This avoids transmission delays caused by multi-layer rendering and system data forwarding, significantly shortening the response time of the screen warning image and ensuring that visual reminders take effect instantly. At the same time, dedicated display control parameters are configured for different target risk levels. The display driver control terminal can precisely adjust the screen imaging quality to match the warning intensity with the level of danger in the scene. This avoids visual interference caused by excessively bright screens in low-risk scenarios and maximizes the visual prompting effect in high-risk scenarios. Combined with a low-latency direct transmission link, this allows the driver to capture the risk warning and perform takeover operations immediately, effectively shortening the driver's reaction time and improving driving safety in human-machine co-driving scenarios.

[0095] As an optional implementation method, determining the display control parameters corresponding to the target risk level includes: If the target risk level is the lowest risk level, then the grayscale parameters are determined, whereby the grayscale parameters are used to enhance the contrast of the image edges; If the target risk level is intermediate risk level, the driving current parameters are determined. The driving current parameters are used to increase the backlight brightness of the area where the road risk object is located and reduce the background backlight to form a contrast between light and dark to enhance the reminder. If the target risk level is the highest risk level, then the color transformation matrix parameters and the mask memory parameters are determined. The color transformation matrix parameters are used to output a single-color warning image, and the mask memory parameters are used to generate a visual funnel containing road risk objects.

[0096] When the processor determines that the current scene is at the lowest risk level, it sends grayscale parameters to the display driver control terminal, such as to the timing controller (TCON). The timing controller is an upper-layer image quality and timing scheduling chip deployed on the motherboard side, receiving display control parameters and image data streams from the vehicle's main controller. These grayscale parameters are used to modify the gamma look-up table (LUT) within the timing controller, relying on a local adaptive contrast enhancement algorithm to achieve nonlinear gamma mapping, only improving the contrast of high-frequency edge details in the image, without changing the overall basic brightness of the image. At the same time, the processor synchronously adjusts the RGB gain of the color management module, smoothly shifting the white point of the image along the Planck trajectory to the 4500K warm color range, completing a slight color shift; and using the two-dimensional bounding box coordinates of road risk objects output by the advanced driver assistance system, a low-transparency envelope is superimposed on the driver layer to gently outline the obstacle contour, providing only low-interference, lightweight visual guidance without affecting the driver's normal observation of the road surface.

[0097] When the processor determines that the current scene is at an intermediate risk level, it generates drive current parameters and sends them to the display driver control terminal, such as to the display driver integrated circuit (DDIC). The display driver chip is the underlying pixel and backlight driver chip, used to convert digital image signals into analog drive voltage or current, directly controlling each group of RGB sub-pixels and MiniLED backlight zones on the screen, executing local backlight overdrive control logic. The processor matches the corresponding MiniLED backlight zone according to the two-dimensional coordinates of the road risk object, instantly increasing the backlight drive current of the region of interest (ROI) where the road risk object is located to 120% of the rated peak value, while attenuating the backlight current of the background area by 30%, relying on the hardware backlight power difference to create a physical-level brightness difference. The processor simultaneously modifies the hardware three-dimensional color lookup table, performs forced color gamut compression and background color desaturation processing on the image; and superimposes a 1.5Hz low-frequency pulse width modulation signal on the target ROI backlight zone, making the backlight produce a continuous breathing flicker effect, relying on dynamic brightness contrast to stably attract the driver's attention and provide a medium-intensity, long-lasting risk warning.

[0098] When the processor determines that the current scene is at the highest risk level, the processor simultaneously issues two sets of control instructions: the Color Correction Matrix (CCM) parameters and the mask memory parameters, and executes a complete set of extreme image quality reconstruction logic.

[0099] Write the step lookup table mapping table to complete the grayscale fault-based extreme value processing: force the low grayscale pixels in the 0-64 range of the image to be set to zero to present pure black, and pull the high grayscale pixels in the 192-255 range to be full to pure white extreme value. The middle grayscale range adopts steep linear mapping to reconstruct the real scene into a high-contrast sketch effect and remove all irrelevant visual redundancy.

[0100] Single-channel color purification is achieved by using color transformation matrix parameters. The display driver chip rewrites the color channel gain, suppressing the gain of the green and blue channels to a noise floor range of less than 5%, and only amplifying the red channel to output a pure warning red image.

[0101] The backlight pulse switching logic is activated, forcing the backlight pulse width modulation duty cycle to switch between 100% and 30% in a square wave mode. The operating frequency is locked in the critical visually sensitive frequency band of 2.0Hz to 3.0Hz (preferably 2.5Hz), directly stimulating the optic nerve to produce a startle response.

[0102] The absolute visual funnel mapping is achieved by directly writing the mask memory parameters to the memory: a Gaussian Alpha (alpha transparency) mask layer is generated with the two-dimensional coordinates of the road risk object as the center. The ROI area in the center of the mask is 100% transparent, and the grayscale of the outer area is reduced by 80% and Gaussian blur is superimposed to form a spatial tunnel visual effect, which forces the driver's line of sight to focus on the risk position within 0.1 seconds.

[0103] Core information such as the takeover countdown is assigned to the highest priority independent hardware display layer (OSD, On-Screen Display). Even if the underlying screen is heavily distorted and reconstructed, the key text information remains clearly visible and will not be obscured by the screen effects.

[0104] In this application, differentiated display control parameters are designed for high, medium, and low risk levels. The lowest risk level uses only a slight warm color shift and a light outline envelope for a lightweight warning, without strong visual stimulation, to avoid interfering with the driver's normal observation of road conditions. The medium risk level relies on the partitioned backlight driving current to create a stable difference in brightness and darkness, providing a continuous and moderate visual reminder. The highest risk level uses multiple strong warning methods, including grayscale extreme value reconstruction, a pure red warning background, a strong focusing visual funnel, and critical frequency flickering backlight, to maximize the intensity of visual stimulation. The three levels of warning intensity are matched with the urgency of the danger in the scene, solving the problems of visual fatigue in low-risk scenes and insufficient warning intensity in high-risk scenes caused by a single warning intensity. In addition, all display control parameters are directly sent to the underlying hardware for execution, bypassing the graphics rendering pipeline of the vehicle's operating system. The image quality reconstruction response latency is extremely low, and the hardware control logic at different levels accurately guides the driver's gaze to focus on the road risk object, shortening the driver's reaction time to identify hazards and complete the takeover operation. This achieves a layered and graded visual warning mechanism that balances driving comfort and driving safety in autonomous driving takeover conditions.

[0105] As an optional implementation, after generating the corresponding warning information according to the target risk level, the method further includes: when a driver's active control signal is detected, restoring each display control parameter to its initial parameters by frame-by-frame interpolation at a fixed refresh rate within a preset recovery time.

[0106] After the processor completes the output of the graded visual warning images, it will continuously monitor the driver's active control signals in real time. The driver's active control signals include the driver's operation signals such as turning the steering wheel, pressing the brake pedal, and pressing the accelerator pedal to take over the vehicle.

[0107] Once the processor recognizes the driver's active control signal, it means that the driver has noticed the road risk and has begun to control the vehicle autonomously, and there is no need to continue to maintain warning display control effects such as high contrast, high brightness, solid color, and visual funnel.

[0108] The processor does not instantly switch all display control parameters to their initial values ​​before adjustment. Instead, it sets a fixed preset recovery time and performs interpolation and gradual adjustment of each display control parameter frame by frame at a fixed refresh rate within this recovery time.

[0109] The initial parameters are the original display parameters before the risk warning is loaded on the screen, including the basic grayscale mapping value, standard backlight drive current, conventional color transformation matrix, blank unmasked video memory configuration, and other original screen configurations.

[0110] Frame-by-frame interpolation refers to the processor dividing the difference between the currently effective display control parameters and the initial parameters. Each frame slightly modifies a set of control values, gradually reducing the intensity of special effects such as warning highlights, color shifts, edge enhancements, and visual masking, and gradually restoring the normal image of the vehicle's infotainment screen until all display control parameters completely return to their initial parameters and the screen returns to its normal display state.

[0111] For example, if the screen was previously determined to be at the highest risk level (Level 3), then the grayscale extreme value parameters, red color matrix parameters, visual funnel mask memory parameters, and peak backlight drive current are enabled. When the driver applies the brakes and takes over, the processor detects the driver's active control signal. Within a preset recovery time of 0.8 seconds, the screen interpolates and adjusts the parameters frame by frame at a fixed refresh rate of 60Hz. Each frame slowly reduces the backlight peak current, softens the red channel gain, reduces the Gaussian mask transparency, and gently corrects the grayscale mapping curve. After a gradual transition over several consecutive frames, all display control parameters return to their initial values ​​after 0.8 seconds, and the screen gradually restores the normal driving display without any sudden or glaring changes in the image.

[0112] In this application, after the driver actively takes over the vehicle, the screen's original initial parameters are restored using a frame-by-frame interpolation gradual transition method. This eliminates the processing logic of instantaneous parameter switching, avoiding drastic changes in brightness and color caused by direct switching between warning and normal images, and preventing sudden image changes from causing visual glare or temporary blurred vision for the driver. At the same time, a uniform preset restoration time is set, coupled with a fixed refresh rate to control the gradual transition rhythm, resulting in a smooth and stable image transition while taking into account visual comfort. After confirming that the driver has actively taken control of the vehicle, the warning effect is automatically canceled, reducing the hardware power consumption caused by continuous special image quality rendering and preventing long-term strong warning images from interfering with the driver's normal observation of the road and vehicle information. After the risk is eliminated, the normal display state is smoothly restored, achieving a balance between driving visual experience and the working stability of the vehicle display hardware.

[0113] Based on the same technical concept, this application provides a warning device, such as Figure 3 As shown, the device includes: The determination module 301 is used to determine the remaining takeover time, target motion entropy, and takeover operation inertia index when a takeover warning signal is detected. The target motion entropy is used to indicate the degree of change in the trajectory of the road risk object, and the takeover operation inertia index is used to indicate the difficulty of the target vehicle to perform the avoidance operation. The first generation module 302 is used to generate the target risk level of the target vehicle based on the remaining takeover time, the target motion entropy and the takeover operation inertia index. The target risk level is used to indicate the degree of driving danger faced by the target vehicle before the driver completes the takeover. The second generation module 303 is used to generate corresponding warning information based on the target risk level. The warning information is used to enhance the driver's perception of driving risks.

[0114] Optionally, the determining module 301 is used for: Acquire the initial takeover duration and driver physiological characteristics, whereby the physiological characteristics are used to indicate the driver's level of concentration while driving; Determine the time reduction factor based on the driver's physiological characteristics; The remaining takeover time is obtained by subtracting the time reduction factor from the initial takeover time.

[0115] Optionally, the driver's physiological characteristics include gaze deviation characteristics, fatigue characteristics, and light adaptation characteristics. The determining module 301 is specifically used for: A line-of-sight reduction factor is generated based on line-of-sight deviation features, which are used to indicate the degree of spatial offset of the driver's line of sight from the vehicle's infotainment screen. Generate a cognitive arousal reduction factor based on fatigue characteristics; A light adaptation reduction factor is generated based on light adaptation characteristics, whereby the light adaptation characteristics are used to indicate the pupil dimming delay caused by the difference between ambient light and the brightness of the vehicle screen. The time reduction factor is determined based on the sum of the visual reduction factor, cognitive arousal reduction factor, and light adaptation reduction factor.

[0116] Optionally, the determining module 301 is used for: Obtain the frequency of abrupt changes in the lateral sway angular velocity and longitudinal acceleration of road risk objects; Determine the variance of the lateral oscillation angular velocity; The sum of the variance of the lateral swing angular velocity and the frequency of abrupt changes in the longitudinal acceleration is normalized to generate the target motion entropy.

[0117] Optionally, the determining module 301 is used for: Obtain the equivalent moment of inertia of the steering wheel and the road surface adhesion coefficient; The quotient of the equivalent rotational inertia of the steering wheel and the road surface adhesion coefficient is normalized to generate the inertia index of the takeover operation.

[0118] Optionally, the first generation module 302 is used for: Input the remaining takeover time, target motion entropy, and takeover operation inertia index into the risk prediction model; The remaining takeover time, target motion entropy, and takeover operation inertia index are mapped to the same feature space through the feature encoding layer. The risk probability distribution is obtained by weighted summation of various mapped features through a hidden layer. From the risk probability distribution, the risk level corresponding to the highest risk probability is selected as the target risk level.

[0119] Optionally, the first generation module 302 is specifically used for: Determine the basic weights corresponding to the three types of features: the remaining takeover time, the target motion entropy, and the takeover operation inertia index. Multiple frames of sensor data corresponding to each feature are collected respectively, and the confidence correction coefficient corresponding to each feature is determined based on the fluctuation of the multiple frames of sensor data. Based on the basic weights and confidence coefficients corresponding to each feature, the target weights of each feature are obtained; The risk probability distribution is obtained by weighting and summing the mapped features using the target weights corresponding to each feature.

[0120] Optionally, the second generation module 303 is used for: Determine the display control parameters corresponding to the target risk level; The display control parameters are sent to the display driver control terminal so that the display driver control terminal can reconstruct the image quality of the vehicle screen based on the written display control parameters. The transmission link for sending the display control parameters bypasses the graphics rendering pipeline of the vehicle operating system.

[0121] Optionally, the second generation module 303 is specifically used for: If the target risk level is the lowest risk level, then the grayscale parameters are determined, whereby the grayscale parameters are used to enhance the contrast of the image edges; If the target risk level is intermediate risk level, the driving current parameters are determined. The driving current parameters are used to increase the backlight brightness of the area where the road risk object is located and reduce the background backlight to form a contrast between light and dark to enhance the reminder. If the target risk level is the highest risk level, then the color transformation matrix parameters and the mask memory parameters are determined. The color transformation matrix parameters are used to output a single-color warning image, and the mask memory parameters are used to generate a visual funnel containing road risk objects.

[0122] like Figure 4 As shown, this application provides an electronic device including a processor 401, a communication interface 402, a memory 403, and a communication bus 404, wherein the processor 401, the communication interface 402, and the memory 403 communicate with each other through the communication bus 404.

[0123] Memory 403 is used to store computer programs.

[0124] In one embodiment of this application, when the processor 401 executes the program stored in the memory 403, it implements the warning method provided in any of the foregoing method embodiments.

[0125] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the warning method as provided in any of the foregoing method embodiments.

[0126] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0127] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0128] It should be understood that the terminology used herein is for the purpose of describing particular exemplary embodiments only and is not intended to be limiting. Unless the context clearly indicates otherwise, the singular forms “a,” “an,” and “described” as used herein may also include the plural forms. The terms “comprising,” “including,” “containing,” and “having” are inclusive and therefore indicate the presence of the stated features, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, elements, components, and / or combinations thereof. The method steps, processes, and operations described herein are not construed as requiring them to be performed in a particular order described or illustrated unless the order of performance is explicitly indicated. It should also be understood that additional or alternative steps may be used.

[0129] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A warning method, characterized in that, The method includes: Upon detecting a takeover warning signal, the remaining takeover duration, target motion entropy, and takeover operation inertia index are determined. The target motion entropy is used to indicate the degree of change in the trajectory of the road risk object, and the takeover operation inertia index is used to indicate the difficulty of maneuvering the target vehicle to perform an avoidance maneuver. Based on the remaining takeover time, the target motion entropy, and the takeover operation inertia index, a target risk level for the target vehicle is generated, wherein the target risk level is used to indicate the degree of driving danger faced by the target vehicle before the driver completes the takeover. Based on the target risk level, a corresponding warning message is generated, wherein the warning message is used to enhance the driver's perception of driving risks.

2. The method according to claim 1, characterized in that, Determining the remaining takeover period includes: Acquire the initial takeover duration and driver physiological characteristics, wherein the physiological characteristics are used to indicate the driver's concentration level while driving; The time reduction factor is determined based on the driver's physiological characteristics; The remaining takeover duration is obtained by subtracting the time reduction factor from the initial takeover duration.

3. The method according to claim 2, characterized in that, The driver's physiological characteristics include visual deviation characteristics, fatigue characteristics, and light adaptation characteristics. The time reduction factor is determined based on these driver physiological characteristics, including: A line-of-sight reduction factor is generated based on the line-of-sight deviation feature, wherein the line-of-sight deviation feature is used to indicate the degree of spatial offset of the driver's line of sight from the vehicle screen; A cognitive arousal reduction factor is generated based on the fatigue characteristics described. A light adaptation reduction factor is generated based on the light adaptation feature, wherein the light adaptation feature is used to indicate the pupil brightness adjustment delay caused by the difference between ambient light and vehicle screen brightness; The time reduction factor is determined based on the sum of the line-of-sight reduction factor, the cognitive arousal reduction factor, and the light adaptation reduction factor.

4. The method according to claim 1, characterized in that, Determining the target motion entropy includes: Obtain the frequency of abrupt changes in the lateral sway angular velocity and longitudinal acceleration of the road risk object; Determine the variance of the lateral oscillation angular velocity; The sum of the variance of the lateral swing angular velocity and the frequency of abrupt changes in the longitudinal acceleration is normalized to generate the target motion entropy.

5. The method according to claim 1, characterized in that, The inertia index for determining the takeover operation includes: Obtain the equivalent moment of inertia of the steering wheel and the road surface adhesion coefficient; The quotient of the equivalent rotational inertia of the steering wheel and the road surface adhesion coefficient is normalized to generate the inertia index of the takeover operation.

6. The method according to claim 1, characterized in that, Based on the remaining takeover time, the target motion entropy, and the takeover operation inertia index, the target risk level of the target vehicle is generated, including: Input the remaining takeover time, the target motion entropy, and the takeover operation inertia index into the risk prediction model; The remaining takeover duration, the target motion entropy, and the takeover operation inertia index are mapped to the same feature space through a feature encoding layer. The risk probability distribution is obtained by weighted summation of various mapped features through a hidden layer. From the risk probability distribution, the risk level corresponding to the highest risk probability is selected as the target risk level.

7. The method according to claim 6, characterized in that, By using a hidden layer to perform a weighted summation of the mapped features, the risk probability distribution is obtained, including: Determine the basic weights corresponding to the three types of features: the remaining takeover time, the target motion entropy, and the takeover operation inertia index. Multiple frames of sensor data corresponding to each feature are collected respectively, and the confidence correction coefficient corresponding to each feature is determined based on the fluctuation of the multiple frames of sensor data. Based on the basic weights and confidence coefficients corresponding to each feature, the target weights of each feature are obtained; The risk probability distribution is obtained by weighting and summing the mapped features using the target weights corresponding to each feature.

8. The method according to claim 1, characterized in that, Based on the target risk level, the corresponding warning information generated includes: Determine the display control parameters corresponding to the target risk level; The display control parameters are sent to the display driver control terminal so that the display driver control terminal can reconstruct the image quality of the vehicle screen based on the written display control parameters. The transmission link for sending the display control parameters bypasses the graphics rendering pipeline of the vehicle operating system.

9. The method according to claim 8, characterized in that, The display control parameters corresponding to the target risk level include: If the target risk level is the lowest risk level, then the grayscale parameter is determined, wherein the grayscale parameter is used to improve the contrast of the image edges; If the target risk level is an intermediate risk level, then the driving current parameter is determined, wherein the driving current parameter is used to increase the backlight brightness of the area where the road risk object is located and reduce the background backlight, so as to form a contrast between light and dark to enhance the reminder. If the target risk level is the highest risk level, then the color transformation matrix parameters and the masking memory parameters are determined. The color transformation matrix parameters are used to output a single-color warning image, and the masking memory parameters are used to generate a visual funnel containing the road risk object.

10. A vehicle, characterized in that, The vehicle includes a memory and a processor, wherein the memory stores executable program code, and the processor is used to call and execute the executable program code to implement the method as described in any one of claims 1 to 9.

Images